search

Neurobiologist / EC601-Pulmonary-Embolism

Development of a Deep Learning Model to Diagnose Pulmonary Embolism
3 stars 2 forks source link

Training deep Neural Nets very slow #9

Open Chris210634 opened 4 years ago

Chris210634 commented 4 years ago

One Epoch (one pass through our data) for ReNeXt-50 on V100 GPU takes approx. 3 hours.

This is unacceptably slow.

Copying data to local scratch space helps a little.